Radeon HD 7970 for Rendering?

[kc1]

Hey guys, I have been looking everywhere and doing so much research but, I have not gotten a clear answer on this. I thought this community could give me a legitimate answer: so I am building a new computer and I need it to basically do everything for me (gaming, HD movies, render pretty well, ect). The thing is I do not want to buy a workstation card because I've feel they wont be that great for gaming as they're meant for one thing and only does that one thing well.

 

How well do 2 HD 7970s (combined with a i7 3930K Processor) fair with Rhino and Vray rendering? I tend to use other programs too such as AutoCAD, Sketch-up, Photoshop, Illustrator, Indesign, grasshoppper, and a few more but I think for the most part they will handle these programs just fine (I am most concerned on how well this set up will render my designs). I might start to render in 3ds Max/Vray RT too (as well as explore other programs to render with), but for now I mainly render in Rhino Vray.

 

I am aware that I wont get top of the line quality from the 7970 for rendering, but I am hoping that it will work just fine overall for my needs.

 

I was torn between the 7970 and the Geforce 680 GTX, because of Nvidia's CUDA cores. However I've been reading that the 680's card architecture doesnt fare well for anything other than gaming? Would going with the 680 be a better choice for my rendering/gaming needs? Also reading up around this forum, There are now new beta drivers that allow the 7970 to render in Vray RT?

 

Can someone give me their two cents on this matter? Thanks! Much Appreciated!

Edited January 5, 2013 by kc1

[Dimitris Tolios]

When writing about GPU renderers, insist on writing "GPU rendering", as the amount of people - even professionals - who are under the impression that the GPU has lots to do with production rendering is astonishing and you/we don't want to perpetuate it.

 

VRay RT GPU just got driver compatibility with 79xx - but only in 3DS. Even that is "fragile" as many people have random success with it.

It is surely promising, as 79xx mop the floor with GTX 6xx/5xx in OpenCL, but that's in Adobe / Benchmarks. The "gods" of GPU rendering have no mercy for AMD thus far.

 

VRay RT in Rhino is CPU only. Doesn't care about your GPU at all. I am not aware of a date for VRay for Rhino 2.0 or w/e version that might expand compatibility to Open CL or CUDA GPUs.

 

Rhino is using OpenGL viewports, where AMD usually has the upper hand in smoothness.

 

Dual GPUs work ok for gaming (micro-shuttering issues do piss some gamers), but all 3D modelling programs couldn't care less - none recognizes / utilizes more than one card, and due to driver issues with gaming cards, none actually utilizes more than 30-40% of their potential for viewport acceleration.

Depending on the games you play and whether you use modded textures etc, I doubt that you need more than one 7970 - unless you will be using surround/multiple screens.

 

I game great with a single GTX 670, driving two monitors (one 1440p - where I am gaming on, one secondary 1080p where I might be running youtube or netflix). I think the 7970 can easily do that and more.

Edited January 6, 2013 by dtolios

[kc1]

Thank you Dimitris, Sorry if I was not clear about the topic at hand. I think judging by the research I've been doing, the 7970 will be a bit of a better choice for me. I do game a lot and I do my main rendering in rhino and Vray. I thought I would start to learn other 3d modelling and rendering programs. That is why I was a little concerned when I heard that the newer Nvidia 6-- series actually does worse than the HD 79-- series in anything not really related to gaming (such as 3d Modelling and rendering). Most likely I will just dabble in 3ds Max or export the rhino file into 3ds max to render in Vray RT.

 

My final question is this: I'm aware rendering programs/plug-ins depend more on your CPU than GPU for the most part. I have always depended on Intel rather than AMD, But it recently dawned on me that AMD has a "8-core" option for the price of 200 dollars (ie: AMD FX-8350 Vishera). I was planning on buying a intel i7 3930K or the i7 3970X, but after finding this, I am not sure which to get. Ive heard that these current gen AMD CPU's dont perform as well as advertised. Can someone chime in their two cents on this? thanks for the response again!

Edited January 7, 2013 by kc1

[Dimitris Tolios]

The 3930K is a much much better value for money in comparison to its Xtreme siblings - 3960X and 3970X. Half the price, and single digit % performance increase - that ofc most "enthusiasts" would cover with overclocking the 3930K, and many graphic pros could consider buying a complete i7-quad or FX 8350 node for the difference in CPU price alone.

 

The FX 8350 is a great chip. It is a true 8-core and with a completely scalable architecture (much bigger 16 core chips existed with similar architecture in the Opteron line for years), but the mediocre by today's standards performance of each one of the individual cores lets it down, as most of the applications outside of servers etc, are single threaded.

 

AMD chips do perform pretty good - their shortcomings are that those don't perform as good as you would expect from them being 6 or 8 core CPUs, simply because we don't compare apples with apples.

 

The 3930K is a chip with 6 sandy bridge cores (SB-E), using hyper-threading (HT) and a share amount of cache memory. So comparing with a 4-core SB with HT "apple", is more or less what you would expect - pretty much the same per clock performance in single threads, pretty much 50% faster in multi-threaded apps.

 

So yeah, the FX 8350, being different than intel's SB or IB architectures, cannot compare core per core and clock per clock to them.

It does pull ahead from quad core i5s of similar cost in purely multi-threaded apps, but that's it.

It is not embarrassingly slow. It is still a 4GHz CPU for god shake - I hate it when people over-exaggerate about being let down etc...

Lets face it - unless you have a 3+ year old machine, it won't rock your world. It won't be 2x faster than a quad. It won't be faster than a 3930K that along with a proper mobo, will cost nearly 4x as much! OF COURSE IT IS FASTER, that's why it is not $200 you fools. doh

 

So? When was that different?

This happened once in my 20 years with PCs...when AMD brought us the Athlon X2, and we got from a single core CPU, to a dual core CPU, within one year, with similar cost, at similar core speeds, at nearly half the heat in comparison with the competition (yes, Pentium 4D was FAIL, remember?)...ok, the 1st quads were a wow too, but not as big of a hit.

We just got greedier since. Too bad it took applications those last 6-7 years to start thinking about becoming multi-threaded in anything but rendering and video/audio transcoding. And most of those that did become, are games.

 

As far as GPUs go...

AMD cards atm are "brute force" oriented. The raw computation capabilities in OpenCL is amazing - in some benchmarks more than double that of the GTX 580, which is the fastest consumer/gaming card from nVidia. Yet all this power has to be harnessed and optimized for apps through drivers, which is where AMD is lacking in many ways. In games, tho many report "better looking" results, the 7970 doesn't pull ahead before huge resolutions get involved for games, and for GPU renderers specifically, we have no support...or they try to, still with not universal success - which is a huge shame imho.

Edited January 7, 2013 by dtolios

[kc1]

Sorry For the Delayed reply, but to sum up everything you are trying to say (correct me if I am wrong):

 

AMD GPUs have the raw power and can hold its own in terms of gaming/rendering/ect. The only problem it has is the driver support. So hence in the field of GPU rendering, it is lacking. However the Nvidia cards (such as the 680), though not quite a "Jack of All Trades"; It does have better driver support and hence, it performs better in terms of GPU Rendering and gaming

 

I am still trying to decide between the HD 7970 and 680 GTX if you haven't noticed.

 

I should also ask this final question: Is there going to be a major difference if regardless of which GPU i choose? It honestly doesn't matter to me if one or the other renders a bit quicker, if its only a difference of a few minutes or so.

 

It would be problematic if either the 7970 or 680 was not able to run a program that I was asked to use though. So If the 7970 or 680 was able to run a few more programs at an efficient pace, I guess I have my answer

Edited January 14, 2013 by kc1

[Dimitris Tolios]

The 670/680 is the jack of all trades. Works well all around - without being too hot or loud (like the 580 which is a tad faster, yet much hotter).

 

The 7970 is on paper amazing. Has more than double the OpenCL computation performance of the fastest (gaming) card nVidia offers (the GTX 580), but it either doesn't work with many of the CUDA only GPU accelerated renderers (Octane, MARI etc), or receives poor driver compatibility with the OpenCL enabled renderers (like VRay RT GPU).

 

The whole point of this thread is for 79xx owners to start veryfying that the latest AMD drivers did restore compatibility and are performing as expected. We could use more testimonies I think, before being comfortable suggesting 7950/7970s for GPU renderings.

 

For general 3D work and viewport acceleration I think Radeon cards work usually pretty well.

[kc1]

Hmm Given your response I think I now have my answer.... Though I REALLY wanted to go with the 7970, But I might have to start using Vray RT/Mental ray soon... So I guess the 680 will work on all of these rendering programs...

 

This might also help me finally decide: Since Vray RT now "works" with the 7970, does it blow the 680 out of the water performance-wise in it? or are they basically neck in neck?

 

Its a shame though, If the 7970 had better driver support, it would be a no brainer

Edited January 14, 2013 by kc1

[Valtiel]

Hey guys. Just a head's up, I just posted my own results with the Chaos Group benchmark scene on the 7950 thread here and results are not good. About 17 minutes. I'm never crazy about the popular cynicism around tech sites regarding AMD's driver record but this is a frustrating development. Seems like such a waste for them to have developed such a monstrous compute engine and shrug driver development for practical OpenCL apps beyond the popular benchmarks used by review sites. It took forever to get a working OpenCL client for Folding@Home too and despite the architectural qualities this same card still gets outrun in PPD by older NVidia (even AMD's own HD6900 series) cards in that application too.

 

I can't recommend these cards for compute and VRAY-RT until they actually start stretching their legs.

[kc1]

Thanks for the heads up, you have now pretty much solidified my decision. I appreciate the post!

[allanmartin]

Guys, Im sorry if im resurrecting the thread but I'm on the same dilemma as the OP was.

 

However, since it's been 6 months, I wonder if AMD has made any big changes to their drivers to make it worth purchasing a 7950/7970 over a gtx 760/770.

 

Thanks!

[Dimitris Tolios]

Afaik not much has changed.

The problem is VRay RT GPU coding imho...it wasn't properly re-adapted for OpenCL, thus tho it works interchangeably with nVidia in either CUDA/OpenCL, AMD performance suffers.

 

Since this "incompatibility" is limited to Chaosgroup's engine, and not any other popular OpenCL accelerated renderer, Adobe CC etc, ppl tend to believe that VRay RT was badly re-written, maintaining some CUDA optimizations/commands vital for fast raytracing.

 

How else could it be that in most OpenCL benchmarks & Renderers in the past, and Adobe CC now, AMD's GCN architecture has like double the performance of competitively priced products, while in VRay RT it is like 5-10 times slower? Did AMD get in bed with all those companies and devs, but left Chaosgroup out? (what a b...).

 

At any rate, there are more reasons to get an AMD card over nVidia since Adobe CC with OpenCL acceleration launched, but Vray RT is not one of them. Lets hope they will catch up with their next version...but I don't see it anytime soon...

[Valtiel]

Hi Allan,

Probably best to go with the GTX. As much as I like AMD's product otherwise, they haven't really made any concerted efforts to resolve this and, from what I now hear from others, VRAY RT was originally written with a CUDA codebase that seems to affect even the OpenCL client. While the HD7000 series is a beast when it comes to compute, I'm beginning to think we'll never see it run RT. Perhaps RT 3.0 will sort this out.

 

Riley

[allanmartin]

Hey thanks guys!

 

The thing is I just started with CG and I dont actually know what Vray RT stands for and it differs from the regular Vray.

 

So far Ive only used the regular Vray for my renderings.

 

Im guessing vray RT is for when you want gpu rendering?

 

Should I be using vray rt at all?

[Dimitris Tolios]

You don't need RT per se. RT is like an "active shade" preview window, which lacks some features, yet it is good enough for many uses.

Given enough time and/or enough GPU power (multiple cards) you can get pretty decent VRay renders in fraction of the time you would get from the CPU. Even single high-end gaming cards (570/580/670/680/780 etc) provide enough compute performance to surpass dual-Xeon systems.

That said, there are still limitations, missing features, driver compatibility and ofc AMD "incompatibility".

 

The "whine" and puprpose of this thread is to just complain for the fact that despite claiming being OpenCL compatible, VRay RT GPU is not very much so, thus the best OpenCL compute cards atm - those based on the GCN architecture by AMD like the 79xx Radeons and W5000/7000/9000 Firepros are virtually useless for Vray RT.

 

But there is no "need" to use it. If you can use Vray, you can easily adapt to using RT CPU or RT GPU accelerated when your hardware allows to do so.

[allanmartin]

What is the average rendering time difference between vray and vray rt with gpu rendering?

 

I really wanted to go with the AMD, probably keep on with the Vray Adv too. But if it will save a huge amount of time maybe I could go with the GTX + Vray Rt.

 

What do you think?

[Dimitris Tolios]

There are some older tests that indicate 15-25x faster render times...no joke.

That is using a i7-920 CPU (4cores/8threads) vs. a GTX480.

 

Issue is, that the AMD Radeons are having such a hit using VRay RT GPU, that it is barely faster than using VRay RT CPU, so it beats the puprose.

 

Now...depending on what you do, VRay RT cannot do 100% of what the CPU VRay Adv does...certain lighting systems, displacements, complex materials, effects etc don't work.

On the other hand, certain camera features like DOF work with very little performance hit (by comparison).

 

 

So it depends on what you want to do, and ofc which is your budget. Vray RT is not the silver bullet for everything.

Yes, atm the best solution for Vray RT is a GTX card. Preferably one with decent VRam buffer @ 3~4GB, though say typical scenes will probably not need as much, it is better to have more than less what's needed. Vray RT simply doesn't compute in the GPU if it cannot fit the whole scene + materials etc in the card's buffer.

[Valtiel]

Aye, I think you'll almost always do your production renders in Adv. I expect that those familiar with the supported features of RT are using it to rapidly test the general look of material, lighting, and compositional changes, speeding up the fine-tuning stages of a scene/animation. One would be able to set up and build their scenes such that they can switch on/off the unsupported features when they come to need this fast draft-turnover. Multiple Titans, afaik, are the ideal cards on an infinite budget for this application.

 

In short, I'd say that technically RT in the right flow can save you human hours on the production of a scene but I believe most would stick with farms and Adv for production render.